Timeline for How to give exercises when students can use ChatGPT
Current License: CC BY-SA 4.0
30 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 27, 2023 at 15:30 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 702 characters in body
|
| Mar 27, 2023 at 15:02 | comment | added | Mahdi Majidi-Zolbanin | @XanderHenderson My answer is saying: don't worry about ChatGPT, it is so bad that, if you give a bunch of random exercises to your students and tell them they have the option to use ChatGPT, those who don't use it will do a better job than those who copy ChatGPT's answers. I was actually more impressed by ChatGPT when I posted my answer, compared to now. But the more I tried it, the more I realized how bad it was. | |
| Mar 27, 2023 at 12:46 | comment | added | Xander Henderson♦ | How is your answer responsive to the question "What should be done about ChatGPT in the learning environment?" | |
| Mar 26, 2023 at 22:45 | comment | added | Mahdi Majidi-Zolbanin | @XanderHenderson "The point is that generative AI is a rapidly evolving field. Any strategy which relies on what it cannot do today is doomed to failure, as we don't know what it might be able to do tomorrow." I do not disagree with that. But that's not news to me. We could make the same statement twenty years ago. What we see today, is pretty bad. In fact very bad. It is so bad that it cannot tell 2 and $\pi$ are not equal, before stating that they are equal. Let me know if you want screenshots to believe me. Yes, we have known for decades that one day AI will take over, but this isn't that | |
| Mar 26, 2023 at 20:14 | comment | added | Xander Henderson♦ | @MahdiMajidi-Zolbanin The point is not that ChatGPT has or has not learned the answer to a particular question. The point is that generative AI is a rapidly evolving field. Any strategy which relies on what it cannot do today is doomed to failure, as we don't know what it might be able to do tomorrow. | |
| Mar 26, 2023 at 18:41 | comment | added | Mahdi Majidi-Zolbanin | in its algorithm, sometimes it gives a correct answer and sometimes it doesn't. But another implication of my answer is this: for what value of n should we start trusting the answers that GPT-n gives us? I have given many examples of incorrect answers by ChatGPT. Of course these are questions that we know the answer to. But what if we don't know the answer to a question and we ask GPT-n? Are we going to believe the answer it gives? What it the criteria for truth here? I can always find questions that GPT-n cannot answer correctly. Or can't I? How do we know I can't? | |
| Mar 26, 2023 at 18:37 | comment | added | Mahdi Majidi-Zolbanin | @XanderHenderson Thanks for your comment. Last night it could not answer correctly whether a triangle and a circle can have the same areas. I tried to teach it and gave an example. Then I asked if it learned it. It said yes, it had "updated its knowledge based on analyzing the new information". I asked the same question after 10 minutes again, but it gave me the wrong answer! But when I asked the same question today, it gave me the correct answer. There are two possibilities: either GPT can really learn (although updating its knowledge is not immediate) or because of a degree of randomness | |
| Mar 26, 2023 at 17:18 | comment | added | Xander Henderson♦ | Indeed, I just asked if a circle and a rectangle could have the same area. The response is now "Yes, if the rectangle is a square with side length $\sqrt{\pi}r$." Give it time, and I am sure that it will give a better answer. | |
| Mar 26, 2023 at 17:13 | comment | added | Xander Henderson♦ | I have downvoted this answer, because I don't think it really addresses the question being asked. Your answer seems to reduce to "ask questions which you know that GPT can't answer". But, as I pointed out in my answer, generative AI is only going to get better. Trying to ask questions which cannot be answered by generative AI is a losing battle. This is not a sustainable approach to the problem (if one views it as a problem). | |
| Mar 24, 2023 at 20:43 | comment | added | JonathanReez | How do we know the answer given by the teacher is correct? :-) it’s an unsolvable problem in the end. | |
| Mar 24, 2023 at 20:25 | comment | added | Mahdi Majidi-Zolbanin | @JonathanReez Yes, I agree with you, it will become increasingly difficult to find homework problems that AI cannot solve. But the question will remain: how do we ever know the answer given by AI is correct? | |
| Mar 24, 2023 at 20:19 | comment | added | JonathanReez | You are right - GPT-n might be able to solve any problem GPT-4 fails at today. But this also means that finding such homework problems will become increasingly hard and thus such advice is quite short lived. | |
| Mar 24, 2023 at 20:11 | comment | added | Mahdi Majidi-Zolbanin | @JonathanReez Are you suggesting that I should find a question that GPT-4 cannot answer correctly? Well, then GPT-5 will answer it correctly. The point I am making is that for every value of n, one can find questions that GPT-n cannot answer correctly. Now, if a student doesn't know the correct answer to a question and asks GPT-n, then how is the student supposed to know whether the answer given by GPT-n is correct or not? | |
| Mar 24, 2023 at 19:54 | comment | added | JonathanReez | For the last one, GPT-4 says: Yes, a rectangle and a circle can have the same area. <...> To find a rectangle and a circle with the same area, you can simply choose the dimensions of a rectangle and then solve for the radius of the circle that would result in the same area. <...>. So your last example needs an update for GPT-4 or can be removed. | |
| Mar 24, 2023 at 19:34 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 218 characters in body
|
| Mar 24, 2023 at 19:31 | comment | added | Mahdi Majidi-Zolbanin | @JonathanReez I don't think WolframAlpha plugin can help with Example 6. Sure, if you ask ChatGPT to revise its answer it may be able to correct it, but to do that, one has to know the correct answer, to begin with. | |
| Mar 24, 2023 at 19:29 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 218 characters in body
|
| Mar 24, 2023 at 14:22 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
deleted 10 characters in body
|
| Mar 24, 2023 at 14:12 | comment | added | JonathanReez | You should try it with the new WolframAlpha plugin: writings.stephenwolfram.com/2023/03/… (once it’s released to the general public) | |
| Mar 23, 2023 at 11:49 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 332 characters in body
|
| Mar 23, 2023 at 11:25 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
deleted 63 characters in body
|
| Mar 23, 2023 at 3:24 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 64 characters in body
|
| Mar 23, 2023 at 3:14 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 1067 characters in body
|
| Mar 23, 2023 at 2:14 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 62 characters in body
|
| Mar 23, 2023 at 2:03 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 500 characters in body
|
| Mar 23, 2023 at 1:39 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 23 characters in body
|
| Mar 23, 2023 at 1:34 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 23 characters in body
|
| Mar 23, 2023 at 1:20 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
added 545 characters in body
|
| Mar 23, 2023 at 0:55 | history | edited | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |
deleted 22 characters in body
|
| Mar 23, 2023 at 0:35 | history | answered | Mahdi Majidi-Zolbanin | CC BY-SA 4.0 |